Free AI SDR Tools: 109 of 264, Sorted Six Ways

109 of the 264 tools in this directory carry `freeTier: true`, and that field is close to useless.

Not because the count is wrong. Because one boolean is holding at least six different products.

Here is the mirror case, and it is on one vendor's single pricing page. Snov.io describes its Trial as "a renewable plan" whose "free credits and recipients renew every 30 days", and eleven rows further down the same page describes its email warm-up quota as "a non-renewable, one-time free quota". Snov.io knows the difference and says so. Our schema records both as `true`.

Nothing here is a measurement. No account was opened, no credit was spent and no invoice was read. Every figure below is either a count computed from this directory's own records or a sentence read off a vendor's published page on a stated date, and the two are labelled differently throughout.

What this page contains: the census with its working shown, a category breakdown where the top row is 80% and the bottom is zero, a fifteen-row table recording what each free tier grants and whether it renews and whether it exports, the export test that removes about half the list, four errors in our own data, and a ten-minute procedure you can run on your own account before spending a week on the wrong one.

The count, and why the count is the least useful part

109 of 264 is 41.3%. That number is computed in code from the `freeTier` field on every record and recomputed on every build, so it cannot drift away from the data underneath it.

It is also the number every listicle would print, and printing it is where those pages stop. The problem is what happens when a buyer acts on it. A reader who takes 109 at face value plans an afternoon of sign-ups. What they hit instead is a set of doors that look identical from outside and open onto different rooms. One gives you 50 credits every month for as long as you want them. One gives you 50 credits once, ever, and then stops. One lets you build a list and refuses to let you take it out of the building.

Of the 109, 12 have no self-serve path, so the free tier requires a sales call to start. 17 publish no paid price of any kind, so you can begin and cannot find out what continuing costs. 42 carry no `pricingVerifiedOn` date at all, which means nobody here has read that vendor's page first-hand, and the free tier is recorded on the strength of an earlier pass rather than a check.

Subtract nothing yet. Those three groups overlap and the overlap has not been computed. The point is narrower: 109 is a count of a flag, and a flag is not a finding.

Sorting it is the work.

What `freeTier: true` is actually holding

Six shapes came out of the fifteen pages re-read on 2026-08-25. They are not a taxonomy invented for the article. Each one is distinguished by something a vendor states in its own words on its own page, which is why the boundaries between them are checkable rather than editorial.

• A recurring monthly allowance. Hunter publishes "50 credits per month" against a $0 plan. It arrives again next month whether or not you spend it.
• A one-time grant at signup. BetterContact: "you get 50 credits for free when signing up." There is no second 50.
• A time-boxed trial. Snov.io offers "a free 7-day trial of LinkedIn Automation" to "all users on plans Starter and higher", which is a free trial you have to be paying to receive.
• A free plan that cannot export. Snov.io again: "unlimited export on all plans except Trial." You can do the work and you cannot take the work away.
• A vendor stating outright that it has none. LeadMagic's help centre asks itself "Is there a free trial or free credits?" and answers with a price and a refund window.
• A record we have not verified. 42 of the 109 carry no read date, and 20 of those 42 are in voice infrastructure alone.

Shapes one and two are the ones that get conflated, and the conflation is not accidental on the vendor side. Both are described on pricing pages with the same three words.

Only one of them is a plan.

Six vendors grant 50 credits and mean three different things

The re-read produced a coincidence sharp enough to be the whole argument. Six of the fifteen vendors set their free allowance at exactly 50 credits. Seven if you count Apollo, whose trial plans "include 50 credits, 5 mobile credits". The number is identical across the set and it describes three unrelated arrangements.

Fifty that comes back

Hunter's Free plan states "50 credits per month" in the plan card, with email verification priced at half a credit each. Snov.io's Trial grants "50 credits, 100 recipients, and 1 email warm-up slot" and adds, in the FAQ directly beneath, "You can use your Trial plan for as long as you like. Your free credits and recipients renew every 30 days." Renewable indefinitely, in the vendor's own words, read 2026-08-25.

Fifty that does not

BetterContact's FAQ: "Yes! We want you to convince yourself of the power of BetterContact, that's why you get 50 credits for free when signing up." FullEnrich's free card lists "50 credits to start" and "No credit card", then adds a third line that gives the shape away: "Earn additional credits by inviting team members." A plan does not need a referral mechanism to replenish it. A grant does.

Dropcontact prints "50 free credits" on its plan cards and "Get 50 emails for free" in its footer, and prices everything else in euros.

Same number, same words, third arrangement.

Fifty that is a demo of a paid feature

Apollo's 50 sits inside a trial of a plan you selected, not inside a free plan. The vendor is explicit about the exit: "You will have the option to convert to a paying plan or downgrade to our Starter plan which is free forever." So Apollo has both shapes at once, and the 50 belongs to the shape that ends. Three arrangements, one number. If you built a shortlist by searching for the number, you built it out of three different things.

The census, with the working shown

Every figure in this section is derived at build time from `ENTITIES` in this repository. The denominator is 264 records across 16 categories. The numerator is the count of records where `freeTier` is `true`. Neither is sampled and neither is estimated.

Read the first row against the last. Voice infrastructure gives a free tier away four times in five. Dialers give one away zero times in eleven.

Sequencing, which is the category most people mean when they type the query, sits at 4 of 12.

That 28-versus-4 contrast is not a difference in generosity and it is not a difference in confidence.

It is a difference in who signs. Infrastructure sells to an engineer who will not commit before testing, so the free tier is the first step of the sale. A dialer sells to a sales manager through a scheduled demo, and a free tier would let that buyer evaluate without the conversation that closes the deal.

The same split runs inside voice. Voice agent platforms, sold to developers, run 59%. Vertical voice agents and inbound qualification, sold to a practice manager or a marketing lead, run 13% and 11%. The product is nearly the same underneath. The buyer is not.

The last column is the one that should temper the rest. Voice infrastructure has the highest free-tier share and also carries 20 of the 42 unverified rows, which is 71% of everything we have not read.

The category that looks most generous is also the category we have checked least, and those two facts should be held together rather than separately.

Fifteen free tiers, read on the same day

Every row below was read first-hand on the date in the last column, off the vendor's own published page, with the billing toggle state noted where it changes what is displayed. "Renews" means the allowance arrives again without payment. "Export" means the free tier permits data to leave the product. Where a vendor does not state something, the cell says so.

Twelve of the fifteen sit in data and research, which is deliberate. It is the category where the free tier is most often the reason someone signs up and most often the thing that turns out not to do the job, because the job is getting a list out of the product and into a sequencer.

Three cells say "Not readable" or "Not re-verified". Those are not padding. Prospeo's pricing page renders its plan cards in JavaScript and returns nothing to a plain fetch, and guessing at what it would have shown is exactly the failure mode this site exists to point at.

The export test, which removes about half the list

There is one question worth asking of a free tier before any other, and it is not how many credits. It is whether the thing you make inside the product can leave it. A free tier that produces a list you cannot download has given you a demo, not a trial, and the difference costs a week.

Snov.io says it in one sentence

Snov.io's own FAQ, read 2026-08-25: "Premium features like integrations, bulk search, bulk verification, teamwork, email A/B testing, API & webhooks access and export are not available in Trial." And elsewhere on the same page, flatter still: "Snov.io offers unlimited export on all plans except Trial."

That is the cleanest statement of the trap in the whole census, and it is the vendor's. The Trial renews forever, which reads as generous.

It renews forever into a room with no door.

Snov.io even says why, in the same FAQ: "the Trial plan has been created to offer a peek into what Snov.io can do and is not meant to help grow sales."

Store Leads costs $250 to get the data out

Store Leads describes its free account as "a limited preview" and lets you browse. The gap is between its paid tiers, not between free and paid. Premium at $75 a month does not include CSV export. Pro at $250 is described as "a fully featured plan, includes exports". Elite is $450 and Enterprise $950, on a 15% yearly discount, read 2026-08-25.

So the honest entry price for anyone who needs the data in a CSV is not $75 and it is certainly not zero.

It is $250, which is 3.3 times the number on the cheapest paid card. Our own record carries $75 as `entryMonthly`, which is correct by the site's convention and is the wrong number for this buyer, and both of those things are worth saying.

BuiltWith is free forever and freezes what you built

BuiltWith states it plainly: "BuiltWith is free to use for individual site lookups forever." One site at a time, unlimited, no card. The moment you want a list of every site running a technology, you are on Basic at $295 a month, and the FAQ confirms what the paid tier buys: "Yes the reports are exportable as CSVs and Excel Spreadsheets. No data is 'locked' into the tool."

One sentence further on is the one worth writing down: "If you go back to Free we'll 'freeze' your reports for when you upgrade again." That is a vendor documenting, without being asked, exactly what happens to your work when you stop paying. Most do not.

Apollo makes export its own currency

The sharpest version of the export test is Apollo's, because Apollo does not gate export at all. It meters it, on a separate clock. The pricing page's own interface strings name three credit types: email credits, mobile credits and export credits. The FAQ defines the third: "You consume export credits whenever you export a contact outside of Apollo. For example, when you use CSV, CRM, or Person API enrichment and sync the data to any system outside Apollo, like Outreach or Salesloft."

So the headline credit allowance on an Apollo plan is not the export allowance, and a free tier measured in credits tells you nothing about how many rows you can remove. Apollo also publishes a second throttle that is not a credit at all: a record selection limit. Its own FAQ describes the consequence in unusually concrete terms, saying that with the limit in place, "to save 10,000 leads you would need to click a button to select 25 leads and add them, 2,000 times."

Twenty-five at a time, two thousand times, is not a price and is not a feature list.

It is a rate limit expressed as tedium, and it is published by the vendor rather than discovered by us. Anyone testing a free plan on the basis of credit count will not meet it until they have already committed a list to the product.

Four things our own data gets wrong

The sort found errors in this repository, and publishing them is worth more than fixing them quietly. A directory that only tells you when it was right has told you nothing about its error rate.

Persana AI is recorded as having no free tier and has one

Persana AI carries `freeTier: false` in `entities.ts`. Its pricing page, read 2026-08-25, opens with a plan card reading "Free / Accessible to everyone / $0 /mo / No billing / 50 credits", and the feature list under it ends with "Export CSV". It is a recurring free plan with export permitted, which puts it in the most useful of the six shapes, and we had it in none of them.

The same record is wrong twice more. `selfServe` is `false` against a page whose free plan button says "Get started", and `entryMonthly` is `null` against published prices of $68 a month billed annually and $85 a month billed monthly for the same Starter plan. Three errors on one row, all in the direction of understating what the vendor offers.

One more thing was on that page and it is the reason to re-check the row again soon rather than mark it done. A banner above the pricing reads "Our next chapter: We're joining forces with Rox." A free tier survives an acquisition about as reliably as a roadmap does, and this site's schema already draws the line between announced and closed for exactly that reason.

Three tools are recorded `true` for a grant that happens once

Dropcontact, BetterContact and FullEnrich all carry `freeTier: true`. All three grant 50 credits once, at signup, and never again. That is not a false entry in the way Persana's is, because the field genuinely records that something free exists. It is a field that cannot say the thing a buyer needs.

The practical size of the gap: Hunter's recurring 50 a month is 600 credits a year. BetterContact's one-time 50 is 50, ever, a ratio of 12 to 1 against records our schema treats as identical. On a straight comparison table those two rows would sit adjacent with the same tick in the same column.

The fix is a field, not a correction

The correct repair is not to flip four booleans. It is to replace one boolean with a shape, so that "renews monthly", "granted once", "time-boxed", "cannot export" and "none" are distinguishable in the data rather than only in prose. Until that lands, treat every `freeTier` tick on this site as meaning "something free exists" and nothing more, and check the shape yourself against the vendor's page.

LeadMagic publishes that it has none, and it is the more useful answer

LeadMagic's help centre has an article whose question is literally "Is there a free trial or free credits?". The answer, verbatim, read 2026-08-25: "LeadMagic is pay-as-you-go. Create an account, choose a plan starting at $49.99/mo, and get your API key immediately. We offer a 14-day money-back guarantee on your first subscription payment."

It never says the word free about itself. It answers the free question with a price and a refund window, which is a straighter answer than three of the free tiers in the table above manage. The refund terms are published in the same detail. The guarantee covers "your organization's first paid self-serve monthly subscription on Basic, Essential, or Growth" within 14 days, and "One guarantee applies per organization." The exclusions are listed rather than implied: credits already consumed, top-ups and overages, renewals, annual subscriptions and enterprise plans, and "unused time, unused credits, or partial billing periods outside the guarantee".

Compare what those two arrangements actually buy an evaluator. LeadMagic's $49.99 buys 2,000 credits and fourteen days to change your mind, with the whole product open. BetterContact's free 50 buys 50 emails, or 5 phone numbers, once. Snov.io's renewable 50 buys as many months of 50 as you like and will not let you export any of them.

By the only test that matters, which is whether you can find out if the tool works for your list, the tool with no free tier is ahead of the two with one.

That inversion is not visible on any page that counts free tiers.

NeverBounce, and what a re-read cost this time

NeverBounce was on the list to re-read because its free-credit count is stated inconsistently across its own properties: a help article giving a small credit figure, a second grant gated behind a demo with no number attached, and a URL slug carrying a number an order of magnitude larger. Three of the vendor's own pages, three answers to one question.

That could not be re-verified on 2026-08-25, and the reason is itself the finding. Both `neverbounce.com/pricing` and the help centre now return a PerimeterX interstitial titled "Access to this page has been denied", to a plain fetch and to a browser user agent alike. The challenge page's own configuration names the brand behind it, setting its custom logo to a zoominfo.com asset. NeverBounce is a ZoomInfo property, and its published free-credit documentation is now behind a captcha.

So the contradiction stands on the earlier first-hand read dated 2026-08-21 and is not re-confirmed today. It sits alongside a second NeverBounce arithmetic problem recorded on the same date: a billing help article prices five verifications at $1.20, which is $0.24 each and 30 times its own published list rate of $0.008, in the same article that prices 5,000 credits at $400 correctly. Treat both as first-party observations dated 2026-08-21 rather than as current facts. A vendor that publishes three different free-credit counts and one impossible unit price, and then puts the pages behind a captcha, is not a vendor whose free tier you should plan a week around.

Fifty free credits is fifty emails or five phone numbers

A credit is not a unit, and the multiplier that breaks it is the mobile one. Persana states it on the free card itself, with no interpretation needed: "50 credits / 50 emails / 1 email found = 1 credit / 5 phones / 1 phone found = 10 credits." The same free tier is either fifty contacts or five, depending entirely on which field you need.

Four of the nine run a 10x mobile multiplier and two run 5x. Apollo does not multiply at all, it meters mobiles on a second counter, which is why its trial states 50 and 5 as two separate numbers rather than one convertible pool. FullEnrich adds a third rate that nobody expects, charging 3 credits for a personal email against 1 for a work email.

The conclusion for anyone running an outbound motion with a phone leg: a free tier is not a phone-coverage test.

Five numbers will not tell you a hit rate, and hit rate is the only property of a phone database worth knowing. FullEnrich publishes its own find rate as "around 80%" and calls it an estimate that varies by industry and geography, which is vendor-reported and should be treated as a ceiling rather than a floor. Five samples cannot distinguish 80% from 40%. This site has priced a contact with both a verified email and a mobile across the same vendors, and the spread there is 5.5x.

The ten-minute test, run on your own account

This is the procedure. It takes about ten minutes per tool and it is designed to fail fast, because the expensive outcome is not a bad free tier, it is a week spent inside one before finding out.

• Minute 0 to 2: sign up and do not configure anything. Skip the onboarding wizard. Skip the integration prompts. You are not evaluating the product yet.
• Minute 2 to 5: attempt an export immediately. Find any ten records, select them and try to get them out as a CSV. Do this before you build anything. Pass condition: a file lands on your disk with the fields you need in it. If the button is greyed, gated, or silently produces a file missing the email column, stop and record the tool as view-only.
• Minute 5 to 7: open the billing page and read the renewal basis. You are looking for one thing, which is whether the allowance has a date attached. "50 credits" with a next-reset date is a plan. "50 credits remaining" with no reset date is a grant. Pass condition: the billing page states when the allowance next refills.
• Minute 7 to 9: find the credit definition and check the mobile line. It is usually in the help centre rather than on the pricing page. Pass condition: the vendor publishes a per-field credit cost. If a mobile costs 10, divide your allowance by ten before deciding what the free tier can test.
• Minute 9 to 10: check whether export is metered separately. Search the pricing page and the help centre for the phrase "export credit". If it returns anything, your credit allowance is not your export allowance and the two need separate arithmetic.

Step two is the one people skip, and it is the one that decides the outcome. Exporting first feels backwards, because you have nothing worth exporting yet. That is exactly why it works: ten junk records cost nothing to lose, and a thousand carefully filtered ones cost a week.

The threshold across all five steps is strict. A free tier passes only if all five pass.

On the fifteen tools in the table above, Snov.io fails at step two, Store Leads fails at step two, BuiltWith fails at step two, BetterContact and FullEnrich and Dropcontact fail at step three, and Apollo fails at step five. Persana and Hunter are the two that look likely to clear it, and neither has been run through the procedure by us, because that would require opening accounts.

What happens when the allowance runs out mid-task

Month three is where free tiers stop being free, and the mechanism is usually not a bill. It is a job that stops halfway through a list, and what the product does next.

Automatic top-up is a paid feature, which cuts both ways

LeadMagic's auto top-up documentation describes the mechanism exactly: you "set a minimum balance threshold", set a top-up amount, and "when your balance drops below the threshold, LeadMagic automatically purchases more credits". The note underneath is the load-bearing one: "Auto top-up requires an active paid subscription."

So on a free tier, nothing silently charges you.

The job simply stops. That is the good news and it is also the bad news, because a bulk enrichment that halts at record 50 of 400 leaves you with a partially enriched file and no record of where it stopped unless the product wrote one.

Upgrading mid-cycle is not prorated, and that is published

The reflex when a job stops is to upgrade. LeadMagic states what that costs: "Changes apply immediately, your billing cycle resets and you are charged the full new plan price (no mid-cycle proration)." Downgrades run the other way, taking "effect at the end of your current billing period".

Apollo publishes the same asymmetry in different words. "Upgrades (plan or seat count) take effect immediately. Downgrades (plan, seat count) take effect immediately and access will update." And then the sentence that settles it: "Refunds will not be processed for downgrades mid-term." Upward instantly, downward without a refund.

Whether the work survives the drop back to free

This is the question nobody asks at signup and everybody asks in month four. Two vendors in this census answer it in writing. BuiltWith: "If you go back to Free we'll 'freeze' your reports for when you upgrade again." Frozen, not deleted, and inaccessible until you pay again. LeadMagic on cancellation: unused credits roll over on monthly Essential plans and above, capped at two months of allocation, while "Basic, annual billing, and custom deals do not roll over".

Read that rollover rule twice. Choosing annual billing, the option the toggle nudges you toward and the one that looks cheaper per month, is the option that removes credit rollover. The discount and the forfeiture arrive in the same click, and only one of them is on the pricing page.

For the other thirteen tools in the table, what happens to your data on downgrade is not stated publicly. That is not an oversight we can fix by reading harder.

It is an absence, and it is the single largest unpublished fact about free tiers in this category.

Free tiers move, and the move is rarely announced as one

The census is a photograph. Everything in it is provisional, and the practitioner record of what happens when a free tier changes is long, specific and unflattering. The clearest case in the Hacker News archive is item 22192543 (422 points, 255 comments, 30 January 2020), a Tell HN posting the vendor's own notification email verbatim. The operative clause: "The main differences are that we are no longer offering 10,000 free emails or 100 free validations." The email around it opens by saying the change is being made "to more accurately reflect the value users get from the service" and describes the effect as "a modest change".

What makes that thread worth citing is not the outrage. It is comment 22193288, which asks the question the notification did not answer: "I am trying to figure out where 625 comes from? Nowhere in the email or on the pricing page that I can find does Mailgun say how many free outbound emails you get now." The title figure of 625 was derived by a customer dividing an invoicing threshold by a per-message rate, because the new allowance was never published as a number.

That is one person's characterisation and not a finding of fact, and it describes the general case exactly. A free allowance can be replaced by an arithmetic identity, and the arithmetic identity does not look like a change to the free tier. The same thread carries the thing that is silently switched off. Comment 22225607: "The email doesn't even mention that 'email receiving' is switched off as a result of moving me to this plan!" The announcement covered the number and not the capability.

Item 24363397 (284 points, 170 comments, 3 September 2020) records the other failure mode, which is a free allowance consumed by the product's own faults. Comment 24366051: "gitlab runners will occasionally hang and just run for the full 1 hour max, making it very easy to max out your hours... I see this happen at least 2x a month." A vendor product manager replied in the thread by name to deny it was deliberate, which is a useful reminder that the mechanism does not require intent to cost you the allowance.

And item 27044371 (840 points, 439 comments, 4 May 2021) carries the version that costs money. Comment 27044666: "Got charged over $800 for 'experimenting' with a DynamoDB database and forgetting to delete it afterwards... something the nice lady on the other end said as she chuckled: 'This happens all the time.'" Anecdote, not evidence of a rate. It is the shape of the risk on any free tier that sits on top of a metered account rather than beside one.

Why free tiers shrink is answered in the same corpus by somebody who worked on it. Comment 22193303: "I worked at one of these companies a few years ago in the department that dealt with spammers. The volume of attempted spam from these free (and low tier) plans was absolutely staggering." That is one person's characterisation, and it is consistent with the direction every free tier in this category moves over time, which is down.

Which free tier is worth an afternoon

Sort by what you are trying to learn, not by the size of the allowance. The allowance is the least discriminating number on the page.

The general rule that falls out of the census: use a free tier to test fit, and a refund window to test cost. They are different instruments. A free tier is capped somewhere that stops you reaching the volume where price behaviour becomes visible, and a refund window opens the whole product for a fixed number of days at a known price. If you are shortlisting five tools this afternoon, budget the afternoon against the export test rather than against the sign-up count. Three of the five will be answered in under ten minutes each, and the two that survive are the only ones worth loading real data into.

What this census cannot see

Forty-two of the 109 free-tier rows carry no verification date. Twenty of those are voice infrastructure, five are voice platforms, five are writing tools and three are enablement. For those rows, the flag records that a free tier existed when the record was created and nothing about whether it exists today.

Some of that gap is not laziness. Nine vendors in this directory return WAF challenges or disallow us by name in robots.txt, and NeverBounce joined them between 21 and 25 August 2026. A page we cannot fetch is a row we cannot verify, and the correct record for it is a blank date rather than a confident tick.

Three specific things this page does not know. Whether Hunter's Free plan permits CSV export: the comparison matrix has a CSV export row spanning all five tiers and the tick states are drawn as images rather than text, so the cell says what it says and no more. What Prospeo's free tier grants: the pricing page does not state it. What NeverBounce's free credit count is today; its page does not show it.

The number to carry away is not 109

109 of 264 tools publish a free tier, and the census is real, and it is still not the answer to the question people are asking when they type it.

Twelve of the 109 need a sales call to start. Seventeen never say what continuing costs. Forty-two have not been read by anybody here. And of the fifteen read on one day in August, roughly half either cannot export or grant their allowance exactly once.

The useful summary is a rule rather than a count. Ask whether it renews, ask whether it exports, and divide by ten if you need a phone number. Three questions, all answerable from a vendor's own pages in under ten minutes, and none of them answered by the word free.

One more from the same afternoon, because it changed our own record and not only our prose: a tool we had marked as having no free tier had a good one, and three tools we had marked as having one were handing out a single allowance at the door.

The field was wrong in both directions on the same pass. If you are choosing on the strength of a directory, including this one, open the vendor's page before you plan a week around what it says. For the inbound side of this question specifically, the free plans that answer a telephone are a much shorter list than the free plans that claim to.